Target recognition and range detection algorithm based on visible-depth image fusion

It is necessary to obtain the surrounding information by the sensor to realize target identification and distance detection in the unmanned situational awareness task. The sensors used to collect target information can be divided into visible light band sensors such as visible light cameras and non-visible light wavelength sensors such as Light Detection and Ranging (LiDAR) from the optical characteristics. The LiDAR focuses on acquiring target distance and reflection intensity information, but the details are limited and not conducive to human observation. The visible light cameras focus on acquiring well-structured and detailed target information, but it is difficult to obtain distance information. In this paper, a target recognition method based on visible-depth image fusion for heterogeneous sensors is studied to solve the influence of environmental, spatial and heterogeneous data factors on fusion results. The fusion of point cloud images and visible light image data into visual images incorporating distance information is achieved by employing a combination of the RANSAC algorithm-based spatial line intersection method and the least squares method. And target recognition and distance detection methods based on yolov5s algorithm were studied for fusion data. Through the establishment of dataset and the training of deep learning network, the specific target identification and distance determination are realized. The results show that the specific target recognition rate of this method is greater than 94%, and the accuracy is 2mm (at 10m distance). The results of target recognition and distance determination can be directly sent to the unmanned situation awareness platform to provide a basis for its next decision.